Vol. 12: 203–214, 2010 ENDANGERED SPECIES RESEARCH Published online September 20 doi: 10.3354/esr00301 Endang Species Res

OPENPEN ACCESSCCESS Habitat use and abundance of striped dolphins in the western Mediterranean Sea prior to the morbillivirus epizootic resurgence

Cédric Cotté1, 2,*, Christophe Guinet2, Isabelle Taupier-Letage1, Estelle Petiau2

1Université de la Méditerranée, OSU/Centre d’Océanologie de Marseille, CNRS, Laboratoire d’Océanographie Physique et de Biogéochimie, Antenne de Toulon, BP330, 83507 La Seyne, France 2Centre d’Études Biologiques de Chizé, CNRS, 79360 Villiers en Bois, France

ABSTRACT: Although the striped dolphin Stenella coeruleoalba has a world status of Least Concern, a recent IUCN Red List assessment has proposed that the Mediterranean population be listed as Vul- nerable and stresses the need for an estimate of abundance and distribution. While substantial efforts have been made to study cetaceans in regions of the western Mediterranean Sea (WM), we have lit- tle knowledge of their large-scale distribution and interaction with oceanographic features at various scales. We conducted 18 basin-wide surveys from ferry platforms within the WM from early Septem- ber 2006 to late July 2007 and used spatial modelling to investigate the distribution, abundance and habitat use of striped dolphins. Most striped dolphins were sighted north of the Balearic Islands, where they are closely associated with the counter-clockwise circulation defined by negative absolute dynamic topography. In the Algerian basin, dolphins were mostly found in recent Atlantic water mass. Densities of striped dolphins were also related to high gradients of sea surface tempera- ture and high chlorophyll concentrations. Moreover, we found similar results (1) when the northern and southern data sets were pooled and (2) in the northern area only, where the majority of sightings occurred. These relationships suggest that dolphins were found in dynamic and productive areas generated by mesoscale processes which could create favourable foraging conditions. Despite our relatively small data set, dolphins were found year-round. We estimated the number of striped dolphins in the area between 3° and 6° E to have been 38 600 (95% CI: 25 900 to 53 900) before the resurgence of the morbillivirus epizootic in summer 2007. This is considerably more than the estimate from the 1991 WM survey, which reported a 43% lower abundance, probably affected by the 1990 epizootic. Platforms of opportunity are therefore relevant not only to assess animal ecology but also to monitor population in a given area. Such interanual monitoring is especially important to detect and quantify the response of animal populations, in terms of distribution and density, to environmental stresses.

KEY WORDS: Habitat use · Abundance and distribution modeling · Mesoscale oceanographic processes · Striped dolphin · Morbillivirus epizootics

Resale or republication not permitted without written consent of the publisher

INTRODUCTION out in the northern part of the WM, i.e. the Ligurian Sea including the Pelagos Sanctuary (e.g. Panigada et Knowledge of cetacean distribution in the western al. 2008, Laran & Gannier 2008), and along the Spanish Mediterranean Sea (WM) is not spatially and tempo- coasts (Cañadas et al. 2005, Cañadas & Hammond rally homogeneous, especially with regard to relation- 2008, Gómez de Segura et al. 2006). These surveys ships with the animals’ oceanographic environment. were often undertaken in the framework of conserva- Because most of the data come from visual observa- tion of wildlife populations through establishment of tions, this knowledge is inherently limited in space and marine protected areas. A large-scale survey con- time. Spatially, observations have mostly been carried ducted during summer 1991 still applies as a reference

*Email: [email protected] © Inter-Research 2010 · www.int-res.com 204 Endang Species Res 12: 203–214, 2010

for distribution and abundance of cetacean popula- with the Mediterranean marine environment (Millot & tions in the WM (Forcada & Hammond 1998). Despite Taupier-Letage 2004). the attempts of a few previous studies to study large- Because of their regular and frequent path across the scale patterns such as the seasonal abundance of WM, opportunistic platforms like ferries are relevant to cetaceans in the Ligurian Sea (Laran & Drouot-Dulau this type of study. Spatial modelling, increasingly used 2007), data acquisition is often limited to the summer in cetacean studies (Hedley 2000, Redfern et al. 2006), season (review in Aguilar 2000). Thus no studies have can indeed accurately analyse data from such plat- yet assessed the distribution of cetaceans in the WM at forms (Marques 2001, Henrys 2005, Williams et al. both large and annual scales. 2006). In addition to presenting several advantages, From an IUCN report (Reeves & Notarbartolo di such as mapping animal abundance and investigating Sciara 2006) and its recent updates, striped dolphin relationships between density and environmental Stenella coeruleoalba in the Mediterranean is cur- covariates, the feasibility of this approach and preci- rently proposed to be listed on the IUCN Red List as sion of estimates have been tested by comparisons Vulnerable. The Mediterranean population of striped with conventional line transect analysis (Gómez de dolphin is particularly exposed to high levels of chem- Segura et al. 2007). Moreover, methods based on gen- icals and heavy metals, which have severe effects on eralized additive models (GAMs; Hastie & Tibshirani their reproduction and immune system (Aguilar & Bor- 1990) were reported to predict cetacean densities on rell 1994). Moreover, Reeves & Notarbartolo di Sciara smaller spatial scales than conventional line transect (2006) reported that the association of mortality in analyses (Ferguson et al. 2006), enabling us to investi- pelagic driftnets (thousands of animals per year) and gate in particular the influence of oceanographic the residual effects of a large-scale die-off from an epi- mesoscale features. We focused our effort on straight zootic in the early 1990s provided the main basis for transects, crossing the WM from northern to southern the proposed listing as Vulnerable. The population shores. Sightings of striped dolphins, in situ oceano- was quantified at 117 880 (95% CI = 68 379 to 214 800) graphic data and remote-sensing images were used individuals in the WM immediately after the 1990 die- simultaneously to examine the spatio-temporal vari- off due to a morbillivirus epidemic (Forcada et al. 1994, ability of striped dolphin distribution and its relation- Forcada & Hammond 1998). Information on striped ship with oceanographic conditions. This variability dolphin distribution and abundance is therefore could also give insights into the potential movements needed to assess current estimates and highlight possi- of striped dolphins, for which different seasonal pat- ble population trends. terns have been reported (Gómez de Segura et al. The biological production of the oligotrophic Medi- 2006, Laran & Drouot-Dulau 2007) according to the terranean Sea, from primary production to top predators, region. is influenced at a basin scale by circulation and air–sea A large-scale and high-resolution multidisciplinary interactions. Within this basin circulation, physical phe- study was conducted shortly before the recent resur- nomena such as meanders and eddies affect biological gence of a dolphin morbillivirus epizootic in 2007 activity, implying a high mesoscale spatial and temporal (Raga et al. 2008). We report the pre-epidemic ecolog- variability (Morel & André 1991, Taupier-Letage et al. ical status and abundance of a WM population of 2003). This can lead, locally and episodically, to unex- striped dolphin. pectedly high biological production in the WM (Taupier-Letage et al. 2003). Many previous studies on marine ecology have linked distribution and abundance MATERIALS AND METHODS of marine predators to high biomasses of resources through biological stimulation and/or aggregation (e.g. Observation and environmental data. Observations Guinet et al. 2001). However, most of these studies are were originally planned for a 2 wk period; however, snapshots of a given, and maybe ephemeral, situation of the meteorological conditions and the ship’s schedule the environment, and they are not designed to consider limited this effort. Data were collected during 18 trips dynamic marine processes. In order to accurately de- between 12 September 2006 and 26 July 2007 on a scribe interactions between cetaceans and their environ- ferry crossing the WM from Marseille (France) to ment in such a dynamic oceanographic context, espe- Algiers and Bejaïa (Algeria) (Fig. 1). Each trip was 2 d cially considering the mesoscale phenomenona, it long and consisted of 2 transects (continuous on-effort is crucial to design a relevant mesoscale-dedicated search time): one from the French coast to the north of sampling strategy. This approach needs to include the Balearic Islands (during daytime on Day 1), and the multidisciplinary high-resolution and long-term (at least other from the Algerian coast to the south of the annual) monitoring. Only such a data set would capture Balearic Islands (during daytime on Day 2). The north- the variability in cetacean distribution and its links ern and southern transects never joined in the middle Cotté et al.: Pre-epidemic striped dolphin habitat and abundance 205

Fig. 1. (a) Transects of the ferries in the study area, the western Mediterranean Sea, used as platforms to observe striped dolphins Stenella coeruleoalba. (b) Locations of striped dolphin sightings along the transects part of the basin, avoiding the violation of indepen- and a fluorometer, estimating the phytoplankton chlo- dence of data by surveying the same area on 2 straight rophyll a (chl a) concentration, at a resolution of 1 mea- days. Observations were made from the bridge (25 m surement per minute. Satellite observations of sea sur- high) of the ship, with the naked eye and with binocu- face temperature (SST), chl a concentrations (visible/ lars equipped with reticle and compass. To ensure the ocean color imagery) and sea level (altimetry) were g(0) = 1 assumption of distance sampling (Buckland et used. The NOAA Advanced Very High Resolution al. 2001), a visual angle of 90° centred on the trackline Radiometer (AVHRR; http://noaasis.noaa.gov/NOAA- (ahead of the boat, 45° port, 45° starboard) was sur- SIS/ml/avhrr.html) satellite sensor, measuring SST, is veyed to optimize the detection of all animals directly an extremely efficient tool that we used to track the on the trackline. Although the ship represents a plat- mesoscale features (e.g. Taupier-Letage 2008). Chl a form of opportunity, we ensured this g(0) = 1 assump- images were obtained from Moderate Resolution tion by avoiding bias of the observers. Consequently, Imaging Spectroradiometer (MODIS, http://modis. the effort and sightings were carried out using stan- gsfc.nasa.gov/). Their spatial resolutions range from dardized (one platform) survey methods by the same ~1 to 4 km, and the WM was fully covered within trained observers. The 2 observers relayed continu- 1 daytime period. Owing to cloud coverage, we used ously, on an hourly basis (one observing, one resting) satellite 3 d-composite images corresponding to during daytime. When a striped dolphin school (single environmental conditions of observation data. We also or several individuals) was sighted and identified from used weekly merged products of absolute dynamic 1 other dolphin species (at a maximum 4 Beaufort sea topography (ADT) at ⁄8° resolution (Ssalto/Duacs, state), ship location, time, radial distance and angle to Aviso, http://atoll-motu.aviso.oceanobs.com). Bathy- the sighting, and group size were recorded. Minimum metric data were extracted from the ETOPO2 data- and maximum group size was estimated from the first base (National Geophysical Data Center, www.ngdc. sighting of a group until the group left the observer’s noaa.gov). Temperature and chl a gradients and field of view. The mean of minimum and maximum geostrophic currents were computed as described in group size was then used in the analyses. Meteorolog- Cotté et al. (2007). ical factors that could affect sighting conditions were Modeling dolphin densities. Line transect sampling also recorded. The ferry travelled at an average speed from design-based methods is the standard tool to pro- of ~20 knots. vide information on abundance and distribution in Environmental data were obtained from in situ and cetacean populations (Buckland et al. 2001). An alter- satellite sources. The oceanographic and meteorologi- native technique relevant to surveys that have not ini- cal in situ data were collected at approximately 3 m depth with a system installed on-board the ferry 1 1 Test phase to develop a network to monitor the surface of the (TRANSMED program ). It consisted of a thermo- Mediterranean using ships of opportunity, an initiative of the salinometer, pumping surface water and recording the Commission Internationale pour l’Exploration Scientifique de hydrological parameters (temperature and salinity), la Méditerranée (CIESM). 206 Endang Species Res 12: 203–214, 2010

tially been designed to achieve equal coverage proba- 1 Dx(,)+ 0 = bility is the model-based approach (Hedley et al. 1999, i 1 ⋅ (2) 2µˆ li+1 Marques 2001), in which line transect sampling is com- bined with spatial analysis. Data from such surveys can where µˆ is the effective strip half-width. Because the then be combined when fitting descriptive models of data set available for the analysis was relatively small, heterogeneity in animal density, thus imparting infor- we chose not to incorporate interaction terms into the mation on how animals use their habitat and how pop- GAM models to avoid the risk of overfitting. We then ulations behave over time (Williams et al. 2006). In the multiplied by the estimated clustered size to calculate present study we used a method based on modeling the total dolphin density. Using a gridded data set of distances between detections, called waiting distances the explanatory variables included in the model, abun- (Hedley 2000). This method has theoretical appeal as it dance was estimated by integrating the density surface avoids the subjective choice of cell size and zero- under the area. The non-parametric bootstrap was inflated data sets generated by the classical method, used to obtain 95% confidence intervals for the esti- which consists of dividing transects into separate cells mates of abundance. The trips, not transects, were (Henrys 2005). Waiting distance models were used to used as the resampling units and were sampled 1000 estimate density because in areas of high density the times with replacement. waiting distance between detections is short (Buck- The covariates used in this procedure are remotely land et al. 2004). The waiting distances were modeled sensed environmental variables: SST and its gradient using GAMs (Hastie & Tibshirani 1990) with a logarith- (gSST), chl a concentration and its gradient (gchl a), mic link function of the set of spatial environmental ADT, derived absolute geostrophic velocity (AGV), sea covariates. To handle over-dispersion, a gamma distri- level anomaly (SLA), derived geostrophic velocity bution was used. We used GAMs in our analysis anomaly (GVA) and depth. Moreover, there is no for- because they offer flexibility through a smooth func- mal rule for specifying the maximum number of covari- tion applied to each explanatory variable (Wood 2003). ates that can be included in any analysis, although Smoothing splines were fitted using multiple general- larger samples (number of observations) tolerate more ized cross-validation (mgcv) in R (Wood 2001). The covariates than smaller samples. Potential covariates amount of flexibility given to a model term is deter- are often strongly correlated and so examination of the mined in a maximum likelihood framework by mini- cross-correlation can be a useful basis for eliminating mizing the generalized cross-validation (GCV) score of some combination between the covariates. First, we the whole model. The general structure of the model identified strongly correlated (|rs| > 0.5) predictors by was: estimating all pairwise Spearman’s rank correlation coefficients (cross-correlation analysis). Second, once []=+β glE(ikki )0 ∑ fz ( ) , i = 1,…,n (1) the non-correlated environmental covariates were k identified, models were built for all possible combina- where g is the link function between expected waiting tion of predictors. β distances E(li) and the covariates, 0 is the intercept, fk To obtain µˆ, we estimated the best detection function are smoothed functions of the explanatory covariates, from conventional and multi-covariate (conditions of zki is the value of the kth explanatory covariate in the sightings) distance sampling (MCDS) methods (Mar- ith observation, and n is the number of observations. ques 2001, Thomas et al. 2002) using the software DIS- While the optimal amount of smoothing is automati- TANCE 5.0 beta 5 (Thomas et al. 2002). The perpen- cally determined by mgcv, the decision as to whether dicular distances were right-truncated prior to the to include or drop a model term is not. We adopted the analysis, following the recommendations of Buckland model selection procedure in the framework recom- et al. (2001). The best detection function was selected mended by Wood (2001), essentially based on the low- using Akaike’s information criterion (AIC) and the chi- est GCV score. The saturated (including all remotely squared goodness-of-fit test. The mean cluster size sensed covariates) models were fitted to the data, and was also estimated by DISTANCE to detect school size a term was dropped from the final model if it satisfied bias, i.e. the tendency of detection for large and small the following conditions: for each model term, the esti- schools at the same distance range. The approach fits a mated number of degrees of freedom was examined to least-squares regression of log of school size on the dis- see if it was near 1. If its 95% confidence intervals tance, which yields a slope when bias is present (not included zero across the range of observations, the the case here). term was removed temporarily, to see if the GCV score Since out of a total of 48 sightings, only 7 were dropped. The density surface between observations, made south of the Balearic Islands, we split the data. given by D(xi + 1,0) was then obtained using the The first analysis was realized on the major core of formula: the data, belonging only to the northern area (strati- Cotté et al.: Pre-epidemic striped dolphin habitat and abundance 207

Table 1. Surveyed distance and number of sightings of striped year would suggest that there is no seasonal trend, and dolphin Stenella coeruleoalba in the western Mediterranean especially no marked difference between winter and from early September 2006 to late July 2007 summer periods (Fig. 2). So, despite the relatively small number of sightings, striped dolphins occurred Region On-effort Number of year-round in the northern part of the WM. distance (km) sightings The best-fitting model of detection probability was a North of Balearic Is. 4351 41 half-normal key function with cosine series expansion South of Balearic Is. 3071 7 and adjustment terms. No environmental covariate Total 7422 48 (not even Beaufort sea state) seemed to affect the detection of dolphins. The expected mean cluster size of striped dolphin was 8.8 (95% CI = 6 to 12.8) and the fied data), while the second model analysis was real- effective strip half-width was estimated to be 224 m. ized on the whole data set, i.e. southern and northern All non-correlated explanatory covariates (Table 2) data (pooled data). were included within the models. Based on the lowest GCV score, the selected spatial models exclude the temporal covariate. This also suggests that the densi- RESULTS ties of dolphins did not markedly vary over the year, i.e. there was a year-long occurrence of striped dolphin Throughout the year, most of the sightings of striped at least in the northern area. From the pooled data set, dolphin schools (~85%, Table 1) occurred in the north- the selected model includes ADT (F = 4.40, df = 3.87, ern part of the WM, i.e. north of the Balearic Islands p = 0.006), SST gradient (F = 2.88, df = 2.04, p = 0.040) (Fig. 1). Encounter rates (number of dolphin schools and chl a concentration (F = 2.95, df = 1.48, p = 0.049). per km) in the northern part of the WM throughout the The deviance explained by this model is 50.1%. The shapes of the functional forms for the smoothed covari- )

–1 2.5 ates, conditional on the other covariates being in- cluded in the models, are shown in Fig. 3. To prevent 2 the over-fitting of our relatively small data set, we imposed a threshold of the degree of freedom of the sightkm

–2 1.5 smoothing spline equal to 4. Waiting distance was low, i.e. density was high, in areas with low ADT. Density 1 was also high in areas with high chl a concentration 0.5 and high SST gradients. We obtained similar results from the stratified (northern) data. The selected model 0 also includes ADT (F = 3.82, df = 3.51, p = 0.019), SST 91011121 2 34 5 6 7 8 gradient (F = 3.13, df = 2.08, p = 0.041) and chl a con-

Encounter rate (x10 rate Encounter Month centration (F = 2.61, df = 1.82, p = 0.061). The deviance Fig. 2. Encounter rate of striped dolphins from September explained by this model is 57.5%. The shape of the 2006 to July 2007 functional form for the ADT (Fig. 4) corresponds to the

Table 2. Results of the cross-correlation analysis of environmental variables based on the Spearman’s rank correlation coefficient rs (| rs | > 0.5 in bold) and corresponding significant levels (lower and upper diagonal). The environmental variables are: sea sur- face temperature (SST) and its gradient (gSST), chlorophyll a concentration (chl a) and its gradient (gchl a), absolute dynamic topography (ADT), derived absolute geostrophic velocity (AGV), sea level anomaly (SLA), derived geostrophic velocity anomaly (GVA) and depth. Significance levels set at <0.05, <0.01 and <0.001; NS: not significant

SST gSST chl a gchl a ADT AGV SLA GVA Depth (m)

SST – NS 0.001 0.001 0.001 NS 0.001 NS 0.05 gSST 0.238 – NS NS NS NS NS NS NS chl a –0.604 –0.133 – 0.001 0.001 NS 0.01 NS 0.05 gchl a –0.544 –0.073 0.803 – 0.01 NS 0.05 NS 0.05 ADT 0.580 0.118 –0.431 –0.427 – NS 0.001 NS 0.01 AGV 0.275 0.214 –0.101 –0.132 0.231 – NS 0.001 NS SLA 0.515 0.134 –0.378 –0.352 0.715 0.135 – 0.05 NS GVA 0.243 0.061 –0.257 –0.279 0.251 0.479 0.347 – NS Depth –0.361 0.071 0.354 0.337 –0.372 0.176 –0.022 –0.055 – 208 Endang Species Res 12: 203–214, 2010

1 1

0 0

–1 –1

–2 –2 –3 –20 –10 0 10 20 30 –25 –20 –15 –10 –5 Absolute dynamic topography (cm) Absolute dynamic topography (cm)

1 1

0 0

–1 –1 –2 Fitted function Fitted function –2 –3 2 4 6 8 10 12 14 2 4 6 8 10 12 14 SST gradient SST gradient

1 1

0 0

–1 –1 –2 –2 –3 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 Chlorophyll a concentration (mg m–3) Chlorophyll a concentration (mg m–3)

Fig. 3. Functional forms of the waiting distance for the Fig. 4. Functional forms of the waiting distance for the smoothed covariates included in the generalized additive smoothed covariates included in the generalized additive model from the northern and southern data (pooled data set) model from the northern data (stratified data set) left part of the curve from the pooled data set (i.e. cence data confirmed the previous results of the range = –25 to 0 cm; Fig. 3), while the other forms were model. In both northern and southern areas, dolphins similar and appear to be less affected by the southern were not found in water masses with a specific SST data (Fig. 4). For both data sets (stratified and pooled), range relative to water masses along the transects. depth, SST, geostrophic currents and chl a concentra- However, they were clearly found in higher chl a con- tion gradient failed to enter the model as covariates. centrations in both the northern and southern regions We then compared in situ oceanographic parame- of the WM. ters, especially salinity (the only parameter not avail- The transect carried out in late September (28– able from remotely sensed imagery), at each sighting 29 September 2006) shows sightings of striped dol- with the mean value of the corresponding transect phins in both the northern and southern regions of the when on-effort (Fig. 5). In the south only, dolphins WM (Fig. 6). The in situ oceanographic profiles of the were found in water masses with a lower salinity than surface water from the on-board system depict a typi- those along the transects. In situ SST and chl a fluores- cal summer situation, with colder and saltier water in Cotté et al.: Pre-epidemic striped dolphin habitat and abundance 209

1.5 the north, and warmer and lower salinity water (recent Atlantic water) in the south. However, in both regions there is spatial variability. In the north there is a weak 1 relative minimum of salinity shorewards correspond- ing to the Northern Current’s signature during sum- 0.5 mertime. In the south the amplitude of the variability is also well marked, especially with respect to salinity. Dolphins were sighted in areas where steep changes in 0 SST and high chl a fluorescence (in situ proxy of chl a at sightings concentration) occurred. These parameters underline –0.5 the mesoscale circulation highlighted by high-resolu-

Environmental conditions tion satellite imagery (1 km in space and 0.1°C for SST precision) and in situ data. In the northern transect, –1 Temperature Salinity Chl a sightings were associated with the external border of fluorescence the Northern Current and with the edge of an eddy. In the southern transect, the sighting was associated with Fig. 5. Environmental preference of striped dolphins. Differ- the edge of a coastal Algerian eddy, well marked by ence between the in situ value of sea surface temperature (SST), salinity and chlorophyll a (chl a) fluorescence at time of low salinity, SST variability and a peak in chl a fluores- sighting and the mean (±SD) of the corresponding transect cence. The transect was also located along another in the north (black bars) and south (grey bars) eddy just north of the previously mentioned coastal

44 France 30 Marseille 27 43 24

SST (°C) 21 42 Spain 18 36.5 38.5 40.5 42.5

41 40 18°C 27°C 39 Balearic Islands 40 38

37 Salinity (psu) Latitude (°N) 39 36 18°C 27°C 36.5 38.5 40.5 42.5 8.5 38 8 7.5 7 37 fluorescence (rel. unit)

a 6.5 Bejaïa 6 Algeria Chl 36.5 38.5 40.5 42.5 345 6 7 Longitude (°E) Latitude (°N)

Fig. 6. Satellite sea surface temperature (SST), striped dolphin sightings and in situ surface parameters of the 28–30 September 2006 transect. (a) The color map represents SST from colder (blue) to warmer water (red). White arrows indicate circulation, white squares are dolphin sightings, continuous black lines are on-effort transects and dotted black lines are off-effort (nighttime). Black points on the map draw the SST profile from in situ data collected by the on-board TRANSMED system, using the transect line as the x-axis (the 2 y-axes are perpendicular to the transect line). (b) Dolphin sightings (gray arrows) are also located along the in situ oceanographic profiles of SST (upper panel, the same profile as on the map), salinity (middle panel) and chlorophyll a (chl a) fluorescence (lower panel). Hatched area: off-effort (night) 210 Endang Species Res 12: 203–214, 2010

Fig. 7. Stenella coeruleoalba. Predicted striped dolphin densities (dolphins km–2) for (a) 31 July–2 August, (b) 15–17 August and (c) 31 August–2 September 2007. White areas within the strips indicate cloud coverage eddy. However, no dolphins were seen within this pattern of the reference work on cetaceans carried out oceanographic context, which is relatively similar to in summer 1991, where striped dolphins were also the neighbouring eddy from the in situ profiles. The mostly found north of the Balearic Islands (Forcada & main difference lies in the SST gradient, which is rela- Hammond 1998). The link with ADT suggests that tively low in the offshore eddy. dolphins were mostly found within the mean counter- Distributions, abundances and confidence intervals clockwise circulation in the north part of the WM, lim- were estimated for 3 periods in August 2007 (early, ited to the north by the Northern Current and to the middle and late) in order to test whether our estimates south by the North Balearic front. Despite the fact that are independent of short spatial and temporal scale a correlation exists between the dynamic height and (i.e. local) conditions (Fig. 7). We then compared this SST (Rio et al. 2007; see Table 2), dolphins were mostly with the abundance of striped dolphins during August found in low ADT areas, but they did not seem to tar- 1991, during the large-scale epizootic event (Forcada get colder waters at a large scale. Indeed, since the & Hammond 1998). We estimated a constant abun- northern region represents the major core of the data, dance of 38 600 (95% CI = 25 900 to 53 900), 37 900 the higher densities were not found at the centre of (95% CI = 24 800 to 52 400) and 38 600 (95% CI = counter-clockwise circulation (the highest value of 27 500 to 51 400) striped dolphins for early, middle and ADT; Figs. 3 & 4) but at the edge (range = –15 to late August, respectively, in the area between 3° and 0 cm). We can only compare our temporal results with 6°E. In this strip, predicted densities were substantially regional studies: striped dolphin abundance in the higher in the north. We estimated a mean density of Ligurian Sea was reported to vary seasonally (Laran & 0.38 dolphins km–1 for the northern region area and Drouot-Dulau 2007) and even throughout the summer 0.12 for the southern region. Throughout their range, months (Azzellino et al. 2008a). However, as in the densities of striped dolphin are higher offshore French present study, dolphins were observed year-round in and Spanish coasts and lower close to the coasts, the central Spanish Mediterranean with no detected despite high densities in the Rhône River plume due to seasonal changes (Gómez de Segura et al. 2006). This especially high chl a concentrations. suggests that the seasonal variation in abundance of striped dolphins depends on the particular area of the WM, with some areas used as summering and/or win- DISCUSSION tering areas, while others are used year-round. De- spite a relatively small data set, the year-long occur- Modeling scale-dependent habitat use rence of striped dolphins which we reported south of the Gulf of Lions could be associated with the high The north–south distribution of striped dolphin production available to dolphin prey. Indeed, the high found in the present study confirms the large-scale primary production within this area is especially dri- Cotté et al.: Pre-epidemic striped dolphin habitat and abundance 211

ven by the mixing generated by the strong NW winds Our results show that observed occurrences of (Mistral) (Morel & André 1991, Bosc et al. 2004) and, striped dolphins are linked to the dynamic environ- to a lesser extent, by land runoff from the Rhône, an ment. This link is especially highlighted by the rela- important fertilization process and source of nutrients, tionship with SST gradient, which defined frontal combined with upwelling driven by winds (Estrada zones. Relationships have been documented with tem- 1996). Moreover, the structure of the Northern Cur- poral SST gradients (Littaye et al. 2004) or deviations rent, flowing along the shelf of the Gulf of Lions, from seasonal averages (Azzellino et al. 2008b), and changes markedly seasonally. During winter, the the spatial link between cetacean density and thermal Northern Current is narrower and deeper than in the fronts has been recently investigated (Gannier & Praca summer, and it displays enhanced mesoscale variabil- 2007). Such important changes in SST are found in tur- ity (Conan et al. 1998, Millot 1999). The mixing mech- bulent areas generated by eddies, meanders and fila- anism induced by the intensification of cross-slope ments in both the northern and southern WM. The current between the shelf and the slope waters may transect shown in Fig. 6 illustrates this link with lead to the seaward export of particulate matter (Dur- mesoscale processes. The salinity is homogeneous in rieu de Madron et al. 1999), which may then be avail- the north; however, dolphins seemed to be associated able to dolphin prey during the oligotrophic winter with steep changes in SST. Listed north to south, the season. Besides temporal variability, it is also specu- external (offshore) edge of the Northern Current (Mil- lated that striped dolphins may change their habitat lot & Taupier-Letage 2005), the edge of an eddy and use as result of behavioural changes that may reflect the north Balearic front are all characterized by abrupt changes in the preference for prey items (Azzellino et changes in both SST and salinity. In the south, dolphins al. 2008a). Despite the fact that sightings did not occur were also associated with the edge of an eddy, corre- very close to the coast, dolphins were observed at a sponding here to rapid change in SST and low salinity, range of distances from the coast. Several studies and a peak in high chl a concentrations. Although (Gaspari 2004, Valsecchi et al. 2004, Bourret et al. striped dolphins were found in areas with high chl a 2007) suggest genetic structuring of striped dolphin concentrations, it is very likely that our model overesti- populations and hypothesize subdivision on the basis mates the densities of dolphins in the northernmost of distance from the coast. Because of the relatively coastal area corresponding to the eastern Gulf of Lions. small data set in the present study, we did not test for Here the plume of the Rhône River corresponds to high the differential habitat use of dolphins according to chl a concentrations, but this area is too close to shore distance from the coast. to present such high densities of striped dolphin, a spe- Relationships between animal densities and covari- cies which is known to be more pelagic than other del- ates at relatively high spatial and temporal resolutions phinids such as common and bottlenose dolphins (few km, days) allow us to interpret these interactions (Cañadas et al. 2002). The mesoscale habitat use of in the context of mesoscale dynamics. Since mesoscale striped dolphins is fairly similar to that of fin whales dynamics modulate the oceanographic and biological reported from concurrent observations and satellite environment, it is important not to spatially and tempo- tracking (Cotté et al. 2009). Since the frontal zones rally average the data over a long period and wide defined by steep changes in SST seem to be important areas, because this would result in smoothing out the for striped dolphins, our results suggest that the phenomena. This would be the case, for example, for mesoscale dynamics create favourable foraging condi- transient upwelling cells induced by strong northerly tions. Although mobile and transitory, mesoscale pro- wind events (few days) in the Gulf of Lions, or for the cesses could potentially present high production, eddies propagating (3 to 5 km d–1) in the Algerian sub- attracting and aggregating both cephalopods and basin (e.g. Puillat et al. 2002), which cannot be identi- mesopelagic fish, the dominant prey species of striped fied on monthly to annual composite satellite images. dolphins (Blanco et al. 1995, Würtz & Marrale 1993, Previous studies (e.g. Cañadas et al. 2002) in coastal Astruc 2005). Such influence of the mesoscale circula- areas have usually reported that cetacean densities tion on the distribution of production and on different were mainly related to static physiographic parame- trophic levels including megafauna has been reported ters, i.e. depth and slope, while dynamic environmen- within the Algerian basin (Viale & Frontier 1994). The tal processes are more rarely investigated. The open presence of aggregated prey is not necessarily the con- sea is characterized by high environmental variability sequence of high chl a concentrations but concomitant as a result of the mesoscale dynamics, despite rela- outcomes of oceanographic processes leading to con- tively constant depth (centre of the basin). It also vergence at frontal zones and favourable conditions for emphasizes the difficulty in examining relationships phytoplankton growth (Franks 1992). The lack of asso- between environmental parameters and cetacean dis- ciation with geostrophic currents as reported for fin tribution from a temporally pooled data set. whales (Cotté et al. 2009) could be due to the strong 212 Endang Species Res 12: 203–214, 2010

influence of advection on krill distribution, while these points are within the range of the other points mesopelagic cephalopods and fish are mobile prey. and they do not severely change the function trend Frontal zones in the open sea have been reported to be (Figs. 3 & 4) linked with other teutophageous predators such as The aim of the present study was not to provide a sperm whales in the WM (Gannier & Praca 2007). new assessment of the WM striped dolphin population, but to show the relevance of the platform of opportu- nity to assess animal ecology and abundance in a given Opportunity to monitor abundance area. Despite different methodologies adapted to ded- icated and non-dedicated surveys, we cautiously com- In their review, Redfern et al. (2006) claimed the rel- pared abundance and density data from our analyses evancy of data sets obtained from platforms of oppor- with previous studies. Our estimate of abundance and tunity for cetacean-habitat modeling purposes. They its confidence interval, in the area between 3° and 6°E, particularly identified the main limitation of oppor- seems substantially higher (by 43%) than the 1991 tunistic data as the variability in the quality and relia- estimate of a mean abundance of 22 100 dolphins from bility of the observations (e.g. the expertise of reported densities in areas north and south of the observers). In the present study, we paid particular Balearic Islands (Forcada & Hammond 1998). Despite attention to this limitation through a rigorous and sys- being more abundant than in 1991, dolphin densities tematic survey protocol applied by the same trained reported in the present study from the northern region observers during all transects. Moreover, our data col- (0.38 dolphins km–1) are slightly lower than densities lection was broad enough to catch spatial and tempo- reported more recently in the Ligurian Sea (0.56, Gan- ral interactions from large to fine scales, drawing a nier 1998) and in the central Spanish Mediterranean more complete picture of striped dolphin ecology than (0.49, Gómez de Segura et al. 2006). However, our den- considering only one scale (Hunt & Schneider 1987, sities are higher than those reported recently in the Redfern et al. 2006, Weimerskirch 2007, Cotté et al. Ligurian Sea from winter aerial (Panigada et al. 2009) 2009). Despite the care taken with the protocol, the and summer boat (Lauriano et al. 2009) surveys. The platform itself can induce inherent bias. The most differences in abundance between the reference study important bias is linked to the assumption that animals of Forcada & Hammond (1998) are higher for the at zero perpendicular distance were detected with cer- northern region (where these authors reported a mean tainty. This so-called g(0) = 1 assumption can be under- density of 0.23) than the southern region (0.08). Actu- estimated in 2 ways: not all animals were available to ally, abundance of striped dolphins from the 1991 sur- be detected (availability bias), or observers failed to vey was conducted after an epizootic mass mortality detect all available animals (perception bias, Marsh & (Raga et al. 2008). A resurgence of the epizootic caused Sinclair 1989). So the g(0) = 1 assumption is valid for by the dolphin morbillivirus seems to have occurred, as highly visible and frequently surfacing animals, there are new die-offs recorded from Spanish and detectable by observers. Gómez de Segura et al. (2006) French coasts. The transects carried out from Septem- report an availability bias for small schools, i.e. a mean ber 2006 to late July 2007, therefore, give the status of proportion of time spent at surface estimated as 0.650 the central part of the striped dolphin population prior (SE = 0.1628), while there is none for large schools, i.e. to the 2007 epizootic morbillivirus event (Raga et al. there is always an individual surfacing. Because the 2008). We do not have enough information on the ferry speed was relatively high (~20 knots), the esti- striped dolphin life cycle to definitively interpret the mated strip half-width was covered in ~22 s, one-third comparison between abundance during the first of the average dive interval of small schools (Gómez de reported epizootic episode in 1990 (Forcada & Ham- Segura et al. 2006). Thus we can roughly estimate a mond 1998) and our pre-resurgence abundance. How- ratio of available animals of ~0.8 for small schools, and ever, the increased density observed between 1991 a ratio of 1 for large schools. Consequently, the avail- and 2007 suggests that the population of striped dol- ability bias was likely to have resulted in minor under- phin may have recovered, at least in part, from the estimation of absolute abundance. Another possible 1990 epizootic. bias in abundance estimates could be due to the small The difference in abundance between 1991 and sample size in the use of GAMs (Hastie & Tibshirani 2007 appears especially marked north of the Balearic 1990). The link between waiting distances and envi- Islands, where densities of striped dolphins were ronmental covariates is based on functional relation- higher compared to south of the islands. This supports ships. The presence of outliers can thus have spurious the hypothesis that epizootics are density-dependent effects. However, the influence of the few data points phenomena (Black 1991), i.e. high population density on the right-hand edge of the distribution should be is likely to facilitate the emergence and propagation of limited, since the waiting distances determined by morbillivirus infections like the 1990 outbreak. In order Cotté et al.: Pre-epidemic striped dolphin habitat and abundance 213

to assess precisely the effects of this new epizootic Cañadas A, Hammond PS (2008) Abundance and habitat event, similar survey protocols from platforms of preferences of the short-beaked common dolphin Delphi- opportunity should be carried out. In the current con- nus delphis in the southwestern Mediterranean: implica- tions for conservation. Endang Species Res 4:309–331 text of changes in the marine environment, it is partic- Cañadas A, Sagarminaga R, Garìa-Tiscar S (2002) Cetacean ularly important to investigate whether the effects of distribution related with depth and slope in the Mediter- environmental stresses, directly or indirectly caused by ranean waters off southern Spain. Deep-Sea Res I 49: human activities, could precipitate the epizootics. 2053–2073 Cañadas A, Sagarminaga R, de Stephanis R, Urquiola E, Hammond PS (2005) Habitat preference modeling as a conservation tool: proposals for marine protected areas for Acknowledgements. We are grateful to SNCM and the ferry cetaceans in the southern Spanish waters. Aquat Conserv crews for their help, and for the contribution of CIESM, Sea- 15:495–521 Keepers and Maritech to the TRANSMED on-board system. Conan P, Pujo-Pay M, Raimbault P, Leveau M (1998) Variabil- The Physical Oceanography DAAC (Jet Propulsion Labora- ité hydrologique et biologique du Golfe du Lion. II. Pro- tory, NASA), Ocean Biology Processing Group (NASA), CLS ductivité sur le bord interne du courant. Oceanol Acta Space Oceanography Division (support from Cnes) and 21:767–782 Météo France are thanked for the distribution of satellite Cotté C, Park YH, Guinet C, Bost CA (2007) Movements of images. C.C. was funded by a Provence Alpe Côte d’Azur foraging king penguins through marine mesoscale eddies. regional council fellowship. Proc R Soc Lond B 274:2385–2391 Cotté C, Guinet C, Taupier-Letage I, Mate B, Petiau E (2009) Scale-dependent habitat use by a large free-ranging LITERATURE CITED predator, the Mediterranean fin whale. Deep-Sea Res I 56: 801–811 Aguilar A (2000) Population biology, conservation threats, Durrieu de Madron X, Radakovitch O, Heussner S, Loye-Pilot and status of Mediterranean striped dolphins (Stenella MD, Monaco A (1999) Role of the climatological and cur- coeruleoalba). J Cetacean Res Manag 2:17–26 rent variability on shelf-slope exchanges of particulate Aguilar A, Borrell A (1994) Abnormally high polychlorinated matter: evidence from the Rhône continental margin (NW biphenyl levels in striped dolphins (Stenella coeruleoalba) Mediterranean). Deep-Sea Res I 46:1513–1538 affected by the 1990–1992 Mediterranean epizootic. Sci Estrada M (1996) Primary production in the northwestern Total Environ 154:237–247 Mediterranean. Sci Mar 60:55–64 Astruc G (2005) Exploitation des chaines trophiques de Ferguson MC, Barlow J, Fiedler P, Reilly S, Gerrodette T Méditerranée par les populations de cétacés. EPHE thesis, (2006) Spatial models of delphinid (family Delphinidae) University of Montpellier. Montpellier encounter rate and group size in the eastern tropical Azzellino A, Airoldi S, Gaspari S, Nani B (2008a) Habitat use Pacific Ocean. Ecol Model 193:645–662 of cetaceans along the continental slope and adjacent Forcada J, Hammond P (1998) Geographical variation in waters in the western Ligurian Sea. Deep-Sea Res I 55: abundance of striped and common dolphins of the western 296–323 Mediterranean. J Sea Res 39:313–325 Azzellino A, Airoldi S, Lanfredi C, Gaspari S (2008b) Biologi- Forcada J, Aguilar A, Hammond PS, Pastor X, Aguilar R cal consequences of global warming: Does sea surface (1994) Distribution and numbers of striped dolphins in the temperature affect cetacean distribution in the western western Mediterranean Sea after the 1990 epizootic out- Ligurian Sea? J Mar Biol Assoc UK 88:1145–1152 break. Mar Mamm Sci 10:137–150 Black FL (1991) Epidemiology of paramyxoviruses. In: Kings- Franks PJS (1992) Sink or swim: accumulation of biomass at bury DW (ed) The paramyxoviruses. Plenum Press, New fronts. Mar Ecol Prog Ser 82:1–12 York, NY, p 509–536 Gannier A (1998) Une estimation de l’abondance estivale du Blanco C, Aznar J, Raga JA (1995) Cephalopods in the diet of dauphin bleu et blanc Stenella coeruleoalba (Meyen, 1833) the striped dolphin Stenella coeruleoalba from the west- dans le futur sanctuaire marin international de Méditer- ern Mediterranean during an epizootic in 1990. J Zool ranée Nord-Occidentale. Rev Ecol Terre Vie 53: 255–271 237:151–158 Gannier A, Praca E (2007) SST fronts and the summer sperm Bosc E, Bricaud A, Antoine D (2004) Seasonal and interannual whale distribution in the north-west Mediterranean Sea. variability in algal biomass and primary production in the J Mar Biol Assoc UK 87:187–193352 Mediterranean Sea, as derived from 4 years of SeaWiFS Gaspari S (2004) Social and population structure of striped observations. Global Biogeochem Cycles 18:GB1005, and Risso’s dolphins in the Mediterranean Sea. PhD thesis, doi:10.1029/2003GB002034 University of Durham, Durham Bourret VJR, Macé MRJM, Crouau-Roy B (2007) Genetic Gómez de Segura A, Crespo EA, Pedraza SN, Hammond PS, variation and population structure of western Mediter- Raga JA (2006) Abundance of small cetaceans in waters of ranean and northern Atlantic Stenella coeruleoalba popu- the central Spanish Mediterranean. Mar Biol 150:149–160 lations inferred from microsatellite data. J Mar Biol Assoc Gómez de Segura A, Hammond PS, Cañadas A, Raga JA UK 87:265–269 (2007) Comparing cetacean abundance estimates derived Buckland ST, Anderson DR, Burnham KP, Laake JL, Borchers from spatial models and design-based line transect meth- DL, Thomas L (2001) Introduction to distance sampling: ods. Mar Ecol Prog Ser 329:289–299 estimating abundance of biological populations. Oxford Guinet C, Dubroca L, Lea MA, Goldsworthy S and others University Press, Oxford (2001) Spatial distribution of foraging in female Antarctic Buckland ST, Anderson DR, Burnham KP, Laake JL, Borchers fur seals Arctocephalus gazella in relation to oceano- DL, Thomas L (2004) Advanced distance sampling: esti- graphic variables: a scale-dependent approach using geo- mating abundance of biological populations. Oxford Uni- graphic information systems. Mar Ecol Prog Ser 219: versity Press, Oxford 251–264 214 Endang Species Res 12: 203–214, 2010

Hastie TJ, Tibshirani RJ (1990) Generalized additive models, (western Mediterranean Sea). Remote Sens Environ 112: Chapman & Hall/CRC, Boca Raton, FL 3400–3412 Hedley SL (2000) Modelling heterogeneity in cetacean sur- Panigada S, Burt L, Lauriano G, Pierantonio N, Donovan G veys. PhD thesis, University of St. Andrews, Fife (2009) Winter abundance of striped dolphins (Stenella Hedley SL, Buckland ST, Borchers DL (1999) Spatial model- coeruleoalba) in the Pelagos Sanctuary (north-western ling from line transect data. J Cetacean Res Manag 1: Mediterranean Sea) assessed through aerial survey. 255–264 J Cetacean Res Manag SC-61-SM7 Henrys P (2005) Spatial distance sampling modeling of Puillat I, Taupier-Letage I, Millot C (2002) Algerian eddies cetaceans observed from platforms of opportunity. MSc lifetime can near 3 years. J Mar Syst 31:245–259 thesis, University of St. Andrews, Fife Raga JA, Banyard A, Domingo M, Corteyn M and others Hunt GL Jr, Schneider DC (1987) Scale dependent processes (2008) Dolphin morbillivirus epizootic resurgence, Medi- in the physical and biological environment of seabirds. In: terranean Sea. Emerg Infect Dis 14:471–473 Croxall JP (ed) The feeding ecology of seabirds and their Redfern JV, Ferguson MC, Becker EA, Hyrenbach KD and role in marine ecosystems. Cambridge University Press, others (2006) Techniques for cetacean-habitat modeling. Cambridge, p 7–41 Mar Ecol Prog Ser 310:271–295 Laran S, Drouot-Dulau V (2007) Seasonal variation of striped Reeves R, Notarbartolo di Sciara G (2006) The status and dis- dolphins, fin- and sperm whales’ abundance in the Lig- tribution of cetaceans in the Black Sea and Mediterranean urian Sea (Mediterranean Sea). J Mar Biol Assoc UK 87: Sea. IUCN Centre for Mediterranean Cooperation, 345–352 Malaga Laran S, Gannier A (2008) Spatial and temporal prediction of Rio MH, Poulain PM, Pascual A, Mauri E, Larnicol G, Santo- fin whale distribution in the northwestern Mediterranean leri R (2007) A mean dynamic topography of the Mediter- Sea. ICES J Mar Sci 65:1260–1269 ranean Sea computed from altimetric data, in-situ mea- Lauriano G, Panigada S, Canneri R, Manca Zeichen M (2009) surements and a general circulation model. J Mar Syst Abundance estimate of striped dolphins (Stenella coeru- 65:484–508 leoalba) in the Pelagos Sanctuary (NW Mediterranean) by Taupier-Letage I (2008) On the use of thermal images for cir- means of line transect surveys. J Cetacean Res Manag culation studies: applications to the Eastern Mediter- SC-61-SM25 ranean basin. In: Barale V, Gade M (eds) Remote sensing Littaye A, Gannier A, Laran S, Wilson JPF (2004) The relation- of the European Seas. Springer, Berlin, p 21–28 ship between summer aggregation of fin whales and satel- Taupier-Letage I, Puillat I, Millot C, Raimbault P (2003) Bio- lite-derived environmental conditions in the north-west- logical response to mesoscale eddies in the Algerian ern Mediterranean Sea. Remote Sens Environ 90:44–52 basin. J Geophys Res 108:3245–3267 Marques FFC (2001) Estimating wildlife distribution and Thomas L, Buckland ST, Burnham KP, Anderson DR, Laake abundance from line transect surveys conducted from JL, Borchers DL, Strindberg S (2002) Distance sampling. platforms of opportunity. PhD thesis, University of St. In: El-Shaarawi AH, Piegorsch WW (eds) Encyclopedia of Andrews, Fife environmetrics. John Wiley & Sons, Chichester, p 544–552 Marsh H, Sinclair DF (1989) Correcting for visibility bias in Valsecchi E, Amos W, Raga JA, Podestà M, Sherwin W (2004) strip transect aerial surveys of aquatic fauna. J Wildl The effects of inbreeding on mortality during a morbil- Manag 53:1017–1024 livirus outbreak in the Mediterranean striped dolphin Millot C (1999) Circulation in the western Mediterranean Sea. (Stenella coeruleoalba). Anim Conserv 7:139–146 J Mar Syst 20:423–442 Viale D, Frontier S (1994) Surface megafauna related to Millot C, Taupier-Letage I (2004) The cetacean world as seen by western Mediteranean circulation. Aquat Living Resour 7: physical and biological oceanographers. In: Briand F (ed) In- 105–126 vestigating the roles of cetaceans in marine ecosystems. Weimerskirch H (2007) Are seabirds foraging for unpre- CIESM Workshop Monograph no. 25, Monaco, p 107–116 dictable resources? Deep-Sea Res II 54:211–223 Millot C, Taupier-Letage I (2005) Circulation in the Mediter- Williams R, Hedley SL, Hammond PS (2006) Modeling distri- ranean Sea. In: Saliot A (ed) The Mediterranean Sea: the bution and abundance of Antarctic baleen whales using handbook of environmental chemistry, Vol 5. Water pollu- ships of opportunity. Ecol Soc 11:1. www.ecologyandsociety. tion. Springer-Verlag, New York, NY, p 1–30 org/vol11/iss1/art1/ Morel A, André JM (1991) Pigment distribution and primary Wood SN (2001) mgcv: GAMs and generalized ridge regres- production in the western Mediterranean as derived and sion for R. R News 1:20–25 modeled from coastal zone color scanner observations. Wood SN (2003) Thin plate regression splines. J R Stat Soc B J Geophys Res 96:12685–12698 65:95–114 Panigada S, Zanardelli M, MacKenzie M, Donovan C, Mélin Würtz M, Marrale D (1993) Food of striped dolphin, Stenella F, Hammond PS (2008) Modelling habitat preference for coeruleoalba, in the Ligurian Sea. J Mar Biol Assoc UK 73: fin whales and striped dolphins in the Pelagos Sanctuary 571–578

Editorial responsibility: Daniel Palacios, Submitted: November 25, 2009; Accepted: June 28, 2010 Pacific Grove, California, USA Proofs received from author(s): August 16, 2010